Significance: Internal organ motion is one of the greatest technical challenges for pancreatic cancer radiation therapy. Extensive research has been conducted on intrafractional tumor motion. This research has provided quantification and enabled novel treatment methods to mitigate motion induced adverse dosimetry effects. Compared to widely studied lung tumor motion, improving pancreas treatment accuracy has equal or greater clinical importance where the tumor is surrounded by radiosensitive serial organs, sparing of which will significantly improve our ability to deliver more effective doses to the tumor. However, organ motion management in the pancreatic cancer treatment is severely underdeveloped due to technical challenges related to poor soft tissue X-ray and CT contrast of pancreas. Fiducial markers are not commonly placed and when placed, inadequately describe complicated multiple-organ motion. Innovation: MRI guided radiotherapy has the potential to overcome these challenges utilizing gated radiotherapy based on organ distances instead of a pre-selected breathing phase. However, to enable it for such pancreatic motion management, MRI acquisition speed needs to be increased so dynamic volumetric images can be acquired with sufficient quality for organ delineation. Furthermore, methods to rapidly digest both pre- and during-treatment MRI images for clinical motion management have not been developed. Both individualized motion margin and gated radiotherapy require explicit organ segmentation but manual delineation of the large number of imaging frames is impractical. Automated segmentation tools for pancreas have not been developed but urgently needed. We will acquire accelerated 3D MRI with sufficient quality for motion quantification by exploiting the spatial and temporal coherence of patient anatomy. We will develop a novel manifold clustering constrained dictionary learning (MCDL) method to efficiently segment the MRI images and provide accurate motion assessment for pancreas anatomy. We hypothesize that the improved motion monitoring will result in significantly improved tumor dose and surrounding normal organ sparing. Aims: 1. Develop methods to acquire 3D dynamic images from under-sampled k-space data. 2. Develop a process to auto-segment prospectively acquired accelerated MRI images by optimizing and validating a MCDL method. 3. Quantify dosimetric gains using accurately described pancreatic tumor motion. Test and robustness and deliverability of the proposed gated plans on a MRI-guided radiotherapy machine (ViewRay). Impact: Patients with locally advanced pancreatic adenocarcinoma have a dismal prognosis but the median survival can be significantly improved for patients who have complete local response. Success of the project will define more accurate patient specific motion margins and facilitate gated radiotherapy that significantly reduce critica organ doses, increase tumor doses for greater complete local response rates. Methods developed by this project will also be applicable to other tumors such as the cervical and liver cancer.